Changelog

July 30

• Context compaction for every model. Long conversations no longer hit a wall. As a chat nears its context limit, older turns are automatically summarized so the discussion keeps flowing, across all models from all four providers. Compaction is handled automatically and never billed to your credits.

July 27

• Three more models: Claude Opus 5, Gemini 3.6 Flash, and Gemini 3.5 Flash Lite.
• High-context support. Send much larger prompts to Grok 4.3, Grok 4.5, Gemini 2.5 Pro, Gemini 3.1 Pro, and OpenAI's long-context models (GPT-5.4, GPT-5.4 Pro, GPT-5.5, and the GPT-5.6 family), so big documents and long histories fit in a single request.

July 20

• Model-specific thinking modes. Thinking effort is no longer a fixed Low / Medium / High. Each model now exposes exactly the reasoning tiers it supports, from minimal all the way up to max, so you get the right depth for the model you choose.

July 17

• Prompt caching, where providers support it. Repeated context (system prompt, tools, long documents) is now cached instead of re-processed, so replies are faster and you save a lot of credits on every follow-up turn. Active on all Claude models, OpenAI GPT-5.4 and newer, and Grok 4.3, with Gemini caching automatically behind the scenes. The savings go straight to you, and billing is never worse than before.

July 14

• Connect your tools with connectors. Link external tools straight into your chats through MCP, starting with seven connectors. Grant access per tool and approve sensitive actions inline.

• You are always in control. Every tool has its own permission: allow, ask, or never. Read-only tools run freely, and any action that writes or changes something asks for your approval right in the chat before it happens.

• Nothing runs without your say-so. Set a connector to read-only, tune permissions tool by tool, or block a tool entirely. Write actions pause and wait for you to approve them once, always, or decline.

• Your credentials stay yours. You connect through the provider's own secure login, so NetusAI never sees your password. Access is encrypted at rest, and you can disconnect any connector at any time to revoke access instantly.
• Built for teams. Owners and admins choose which connectors an organization can use, and owners can enforce a read-only ceiling so no member can trigger write actions.

July 13

• The GPT-5.6 family lands. Three new OpenAI models, Luna, Terra, and Sol, spanning cost-efficient to flagship performance.

July 3

• Say hello to xAI. Grok joins the lineup as our fourth provider, alongside Claude, Gemini, and OpenAI.
• Three new models: Claude Sonnet 5, Claude Fable 5, and Grok 4.3.

July 2

• Share a chat. Chats in a project are private to whoever started them by default, even when other members join the project. When you want input, share any conversation with your project members so the whole team stays on the same page. Nothing you create is visible to others until you choose to share it.

July 1

• Move chats in and out of projects. Reorganize freely by moving any conversation into a project or back out to your personal space.
• New "Recent" filter. Jump back to what you were working on with a filter that surfaces your most recent chats first.

June 26

• Per-member credit limits for organizations. Owners and admins can now cap how many credits each member uses over a rolling 30-day window, keeping team spend predictable. Members are notified when they reach their limit.

June 24

• Regenerate any response. Not happy with an answer? Regenerate it in a single click.
• Conversation branching. Regenerating or editing a message now creates a branch instead of overwriting it, so you can explore different directions and move between versions with the ‹ › navigation. Nothing is ever lost.

• Pin your chats. Pin and unpin conversations to keep the ones that matter at the top of your list.

June 23

• NetusAI Chat is here. Chat with the latest models from Claude, Gemini, and OpenAI in one place, with streaming responses, thinking modes, and built-in web search. Group conversations into projects, add knowledge sources for grounded answers, and switch models mid-conversation without losing your place.